Researchers at the Indian Institute of Science (IISc) have developed a new graphics processing unit (GPU)-based machine learning algorithm that could help scientists better understand and predict connectivity between different regions of the brain.
The algorithm, called Regularized Accelerated Linear Fractionation Evaluation, or ReAl-LiFE, can rapidly analyze large amounts of data generated by diffusion magnetic resonance imaging (dMRI) scans of the human brain.
Using ReAL-LiFE, the team was able to evaluate dMRI data 150 times faster than existing state-of-the-art algorithms, according to an IISc press release published Monday.
Devarajan Sridharan, Associate Professor at the IISc Center for Neuroscience (CNS), said: “Tasks that previously took hours to days can be performed in seconds to minutes. Nature Computing Science.
Millions of neurons fire in the brain every second, creating electrical impulses that travel from one point in the brain to another through connecting cables, or “axons,” a network of neurons. These connections are critical to the computations the brain performs.
“Understanding brain connectivity is critical to uncovering the relationship between the brain and behavior at scale,” said Varsha Sreenivasan, a CNS doctoral student and the study’s first author. However, traditional methods to study brain connectivity typically use animal models and are invasive. On the other hand, dMRI scans provide a non-invasive way to study the connectivity of the human brain.
The cables (axons) that connect different regions of the brain are its information highways. Because axon bundles are shaped like tubes, water molecules pass through them in a directed fashion along their length. dMRI allows scientists to track this movement to create a comprehensive map of the brain’s fiber network, known as the connectome.
Unfortunately, pinpointing these linkers is not straightforward. The press release notes that the data obtained from the scans only provided the net flow of water molecules at each point in the brain.
“Imagine that water molecules are cars. The information obtained is the direction and speed of the vehicle at each point in space and time, with no information about the road. Our task is similar to inferring the road network by observing these traffic patterns,” Sridharan said. explained.
To accurately identify these networks, conventional algorithms closely match the predicted dMRI signal from the inferred connectome to the observed dMRI signal.
Scientists previously developed an algorithm called LiFE (Linear Fractionation Evaluation) to perform this optimization, but one of the challenges is that it runs on a traditional central processing unit (CPU), which makes the calculations time-consuming.
In the new study, Sridharan’s team tuned their algorithm to reduce computation in several ways, including removing redundant connections, thereby significantly improving LiFE’s performance.
To further speed up the algorithms, the team also redesigned them to work on specialized electronic chips (of the kind used in high-end gaming computers) called graphics processing units (GPUs), which helped them perform faster than computers. Analyze data 100-150 times faster. previous method.
This improved algorithm, ReAl-LiFE, was also able to predict how human test subjects would behave or perform specific tasks.
In other words, using the algorithm’s estimated connection strengths for each individual, the team was able to explain variations in behavioral and cognitive test scores for a group of 200 participants.
Such analysis could also have medical applications. “Large-scale data processing is becoming increasingly necessary for big data neuroscience applications, especially for understanding healthy brain function and brain pathology,” Sreenivasan said.
For example, using the obtained connectome, the team hopes to be able to identify early signs of aging or the deterioration of brain function before the behavior of Alzheimer’s patients manifests itself.
“In another study, we found that a previous version of ReAL-LiFE did a better job of distinguishing Alzheimer’s disease patients from healthy controls than competing algorithms,” Sridharan said.
He added that their GPU-based implementation is very general and can also be used to solve optimization problems in many other domains.